4 CAN Newsletter 2/2015

Concierge and assistance robot

Hollie is a mobile, bi-manual service robot that was de-veloped at the FZI (Forschungszentrum Informatik)

Karlsruhe as part of the “House of Living Labs” (Holl). In the future, the robot will carry out different tasks within the Holl, for example accompany visitors and provide assis-tance. It comes as no surprise, then, how we arrived at the robot’s name: Hollie is an acronym for “House of Living Labs intelligent Escort“. With the help of speech synthe-sis software, a microphone array and an LED-based infor-mation system within the body, the robot offers possibili-ties for human-machine interaction that go beyond simple information services. New tasks can be commanded to the robot via 3D gesture recognition. This also enables users to teach Hollie new trajectories, as the robot can mimic and store human motion sequences.

During the design phase of the service robot Hollie, the focus was on human-robot interaction as well as the robot’s practical usability for a human user. From these requirements we derived the proportions and kinematic structure of the robot’s body as well as the design of the outer hull with its round and friendly geometries. Wherever possible, we applied robust industrial components in the construction. Accordingly, almost all actuators come from Schunk: Two LWA 4p light-weight robotic arms made with Powerballs with CANopen interfaces are employed as ma-nipulator arms and the neck is also made of a Powerball. The upper body is actuated by two Schunk PRL 120 high-torque rotary modules, which are supported by springs when the body is close to maximum deflection. This is the case when the robot leans forward to reach objects on the floor. The shoulder axis that carries the arms and the neck is kept in a horizontal alignment by a parallelogram system of levers within the body. This parallelogram system also relieves the upper body actuators from the torques gen-erated by the arms when they are stretched out forwards or backwards. Currently Hollie is equipped with two an-thropomorphic servo-electric 5-finger gripping hands (Sc-hunk SVH) that allow maximum flexibility when grasping and manipulating everyday objects.

The mechanical structure of the body allows Hollie to reach the ground with its hands in a way that is simi-lar to humans. Grasping objects from the floor is a major requirement for a robot if it is meant to be a real help for people with physical handicaps. Therefore, Hollie is able to autonomously locate and fetch dropped keys or pieces of clothing. On the other hand, the body kinematic allows the robot to reach high shelves or windows when stretched up. The robot then has a shoulder height of 124 cm, which is sufficient for example to get dishes from an upper kitchen

Assistance robots are still very much in development, but Hollie already has the basics down: it manages complex tasks and supports people in everyday situations. The robot can even reach the floor by bending its body forward.

cabinet and set the table. And that is in fact one of the main application scenarios for the robot: To unburden service staff from repetitious tasks like setting or cleaning the table, for example in an elderly care center, escorting people in public places like in a museum, or execut-ing mobile pick and place tasks in shop-floor logistics applications or industrial environments. For that, Hollie has a mobile platform with an omni-directional drive system from Segway, which is controlled by the central computer via a CAN interface.

Figure 1: Hollie performing an interactive dexterous manipulation task (Photo: FZI)

Figure 2: To relieve elderly care personnel from repetitive tasks, Hollie can set the table (Photo: FZI)

App

licat

ions

Powerful Control Units for High-Safety Applications: HY-TTC 500 Family

www.ttcontrol.com/HY-TTC-500-Family

Flexibility & Usability• Single controller for whole vehicle

for centralized architectures • Extensive I/O set with multiple software

configuration options per pin • Open programming environments C, CODESYS®

V3.x and CODESYS® V3.x Safety SIL 2

Safety • TÜV-certified according to IEC 61508

(SIL 2) and EN ISO 13849 (PL d) • ISO 25119 AgPL d certifiable• CODESYS® Safety SIL 2 including support

for CANopen® Safety Master and easy separation of safe / non-safe code

• Safety mechanisms in hardware to minimize CPU load • Up to 3 output groups for selective

shut-off in case of safety relevant fault • Safety companion and safety mechanism in hardware

Connectivity • Up to 7 CAN interfaces • Automatic baudrate detection and

configurable termination for CAN • Ethernet for fast download and

debugging purpose

Performance • 32 bit / 180 MHz TI TMS570 dual core

lockstep processor (ARM architecture) • Up to 2.3 MB RAM / 11 MB Flash• Floating-point-unit

Robustness • Automotive style housing suited for

very rough operating conditions • Total current up to 60 A

General PurposeECUs

SafetyCertifiedECUs

Safe I/O Modules

Operator Interfaces

I/O Modules

6 CAN Newsletter 2/2015

Two Sick laser range finders with a 270-degree field of view each are mounted on opposing corners of the plat-form, slightly above the wheels. The laser range data is used for navigation purposes as well as for collision pro-tection. A software package that has been deployed in au-tonomous unmanned transportation systems for hospitals for over ten years now allows Hollie to navigate in all en-vironments without requiring any structural modifications or markers. The robot is small enough to maneuver in any building that is suitable for wheelchairs.

The delicate outer appearance of the robot deceives about its weight of 160 kg. Within a footprint of around 80 cm x 80 cm, batteries, two computers, network equipment, sensors, and a total of 61 movable axes are located. Of these 61 axes, 38 are individually actuated.

To achieve an appealing form language of the outer paneling without restricting the functionality of the robot, we made use of large, individually designed 3D-printed parts. Also, the head hull, which acts as a mount for a stereo camera system, an RGB-D sensor, and various

microphones, was completely printed in two parts via Laser

Sintering. The front and back panels of the body are

covered

with fab-ric, as they

have to be flex-ible when the

body is bending. Hollie’s human-like ap-

pearance invites users to interact with the

robot while it is still far enough from

being too human-oid, as this can

be perceived to be creepy – roboticists know this phenomena

as the “Uncan-ny Valley”.

Hollie is able to recognize persons via

face detection algorithms. Com-bined with the anthropomorphic upper

body, this enables a broad field of applica-tion scenarios: Starting with the motivation for physical training of older people (the robot demon-strates Tai Chi exercises), to the usage as an em-bodied telepresence sys-tem (the robot represents the contact person), and last but not least as a mobile nursing assistant (the robot patrols a build-ing, reaches patients by opening doors, and uses elevators).

The energy sup-ply of the robot is built of four lithium-polymer rechargeable batteries, which power the mobile platform and all on-board electronic devices, ex-cept the logic and power stages of the upper body actuators, including the arms and hands. These are fed from an addition-al lead acid gel battery. A switch-over-circuit allows the charging of all batteries during operation and the seamless switching between external and internal power supply while in operation.

The total number of 38 actuators is controlled by an architecture consisting of two computers that are interconnected via Gigabit-Ethernet. The internal network also connects the two laser range finders and a Wifi bridge. A high-performance Linux Quad-Core embedded computer handles the sensor-data process-ing, the navigation, and the motion planning. The second embedded system is a dedicated interpolation computer, responsible for the low level control of all actuators. Therefore, it is equipped with four CAN interfaces: The Segway platform requires two separate CAN networks, the other two buses run a CANopen protocol and control twelve arm nodes plus four additional nodes in the torso and neck modules. Also, two serial EIA-485 buses are required for the two Schunk SVH hands, which each posses nine current-controlled degrees of freedom.

The real-time control and the low-level hardware abstraction layer are implemented in FZI’s own MCA2 robot control framework, written in C++. The event-based high-level software was composed in the open source framework ROS (Robot Operation System) and contains C++ and Python nodes. Both frameworks are network transparent, so that single function blocks can also be run off-board.A detachable tablet computer on the robot’s back as well as a speech-dialogue system can be used to command tasks.

Figure 3: Mechanical structure of the upper body that allows the robot to bend forward (Photo: FZI)

App

licat

ions

Author

Andreas HermannFZI Forschungszentrum Informatik www.fzi.dehermann@fzi.de

Related linksHollie picks up laundry from the floorOver hill, over dale: The six-legged robot does wander everywhere

SENSORS FOR MOBILE MACHINES

Absolute Rotary Encoders and Inclinometers

Reliable Measurement under Harsh Conditions

High Protection Class: IP69K

Fieldbus and Analog Interfaces

Safety and ATEX

Ex-Proof Versions Available

Successfully Integrated in

Concrete Pumps, Drilling Machines,

Working Platforms, Cranes, Wheel Loaders,

Leader Masts and More

POSITALGermany, Singapore and USA

www.posital.com, info@posital.com

www.posital.com

The robot is also equipped with a full color LED-ribbon that runs around the mobile platform and which is used to signal driving intentions or the direction and distance to an obstacle that blocks the task execution.Another full color LED-badge in the chest of Hollie visualizes the overall system status, which includes the battery status or the degree of autonomy.

Recently, the robot was equipped with a 3D collision detection system that runs on a GP-GPU (General Pur-pose Graphics Processing Unit). The system processes and interprets point-cloud data from Hollie’s 3D camera and does a live motion prediction of all moving obstacles in the surroundings. These predictions are taken into con-sideration while planning the robot’s motions. The algo-rithms are fast enough to plan collision-free trajectories in dynamic environments without noticeable delays. Thus, the robot can even be employed in close proximity to humans and safely execute its assigned tasks.

Altogether, Hollie is a flexible and innovative service robot that can be used as a research platform for a manifold of research aspects. Numerous demon-strations and exhibitions on fairs have demonstrated the practical usability and stability of the industrial components as well as the potential of FZI’s robotics software. t

Figure 4: Human-sized servo-electric 5-finger hands from Schunk allow Hollie to grasp and manipulate objects (Photo: FZI)